Accelerating enterprise code editing using AWS with Morph

Morph’s flagship product, Fast Apply, is a specialized small language model that merges code updates from large language models, such as Claude, into original code files at exceptional speeds. Rather than competing with larger tools, Morph provides the middleware that makes those solutions fast, reliable, and cost-effective at scale.

As frontier AI models improve, the bar for competitive performance keeps rising. For Morph’s own needs, what started at 1,000 tokens per second quickly became inadequate. Morph needed to reach 10,000 tokens per second while maintaining first-pass accuracy and stable production performance. Managing memory bandwidth for concurrent users on a single GPU presented optimization challenges; off-the-shelf inference engines didn’t properly allocate bandwidth for stable, high-speed production workloads. Many of Morph’s enterprise customers required secure, on-premises deployment with strict legal and compliance guarantees, and they needed cost-effective single-GPU options rather than expensive multinode configurations that would price out most use cases.

Using AWS, Morph could gain access to the infrastructure, deployment tools, and security architecture needed to train high-performance models at scale and deploy them securely for large organizations. “AWS is infrastructure I can trust,” says Tejas Bhakta, founder and CEO of Morph. “I know AWS is going to be around—AWS has tried-and-tested solutions, and I’m not going to encounter hardware failures or edge cases with memory sharing.”

Morph trained its Fast Apply model using Amazon Elastic Compute Cloud (Amazon EC2) P5 Instances—high performance GPU-based instances for deep learning and high performance computing applications—powered by NVIDIA H100 GPUs. The team built a custom inference engine specifically optimized for code merging tasks, using speculative decoding techniques to maximize speed. “We trained all these models on AWS, and we started at around 1,000 tokens per second, and then kept doing optimizations,” says Bhakta. “We got up to 2,000, and then 4,000, and eventually we reached 10,000.”

“To achieve those speeds reliably in enterprise settings, we engineered a custom inference engine and GPU scheduling layer that maintained deterministic performance even under high-concurrency workloads,” says Dhruv Bhatia, founding researcher at Morph.

After the models were trained, Morph needed a secure way to deploy them to enterprise customers without exposing proprietary model weights. For inference deployment, the company adopted Amazon SageMaker AI, a fully managed service that brings together a broad set of tools to facilitate high-performance, low-cost AI model development for any use case. Morph also listed its models on AWS Marketplace, where customers can discover, deploy, and manage solutions for the cloud. With a single click, customers can deploy Fast Apply directly into their own secure environments. Meanwhile, Morph retains full protection over its intellectual property.

“Amazon SageMaker AI provides a security-provisioned instance where we can deliver our inference engine and weights, knowing they’re protected,” says Bhakta. “That reduces our go-to-market burden; we can publicly list on AWS Marketplace instead of doing bespoke deployments for each customer. Otherwise, we’d need to run containers in their infrastructure and ensure that safeguards and licenses are signed to prevent unauthorized training or access.”

With its models running on AWS infrastructure, Morph achieved over 10,000 tokens per second per request, a major leap from the 1,000 tokens per second where it began. The company can now complete a 15,000-token multifile refactor in under 400 milliseconds, compared with 20 seconds by traditional approaches. Single-file coding edits that previously took 2–5 minutes now finish in under 1 second.

Binance—a major financial services company with 5,000 developers—reported that its teams became 50–70 percent more effective using Morph’s Fast Apply model within their internal integrated development environment. These developers can code much faster than they could previously; large-scale refactors that previously required months now take just days, changing how quickly Binance can evolve its codebase.

For enterprises operating under strict legal and compliance requirements, Morph’s AWS infrastructure made coding agents viable. The combination of Amazon SageMaker AI security architecture and AWS Marketplace distribution built the trust necessary for widespread enterprise adoption. Morph now serves over 20 coding services and continues expanding its enterprise customer base.

Looking ahead, Morph plans to develop semantic search capabilities using embedding and reranking models to make coding agents work effectively on very large repositories. By building on AWS, Morph is positioned to lead the development of specialized models that make AI coding agents production ready at enterprise scale.